22 KiB
Upgrading apps to Corda 4
These notes provide instructions for upgrading your CorDapps from previous versions. Corda provides backwards compatibility for public, non-experimental APIs that have been committed to. A list can be found in the corda-api
page.
This means that you can upgrade your node across versions without recompiling or adjusting your CorDapps. You just have to upgrade your node and restart.
However, there are usually new features and other opt-in changes that may improve the security, performance or usability of your application that are worth considering for any actively maintained software. This guide shows you how to upgrade your app to benefit from the new features in the latest release.
Step 1. Adjust the version numbers in your Gradle build files
Alter the versions you depend on in your Gradle file like so:
ext.corda_release_version = '4.0'
ext.corda_gradle_plugins_version = '4.0.38'
ext.kotlin_version = '1.2.71'
ext.quasar_version = '0.7.10'
Note
You may wish to update your kotlinOptions to use language level 1.2, to benefit from the new features. Apps targeting Corda 4 may not at this time use Kotlin 1.3, as it was released too late in the development cycle for us to risk an upgrade. Sorry! Future work on app isolation will make it easier for apps to use newer Kotlin versions than the node itself uses.
You should also ensure you're using Gradle 4.10 (but not 5). If you use the Gradle wrapper, run:
4.10.3 ./gradlew wrapper --gradle-version
Otherwise just upgrade your installed copy in the usual manner for your operating system.
Step 2. Update your Gradle build file
There are several adjustments that are beneficial to make to your Gradle build file, beyond simply incrementing the versions as described in step 1.
Provide app metadata. This is used by the Corda Gradle build plugin to populate your app JAR with useful information. It should look like this:
cordapp {4
targetPlatformVersion 4
minimumPlatformVersion
contract {"MegaApp Contracts"
name "MegaCorp"
vendor "A liberal, open source licence"
licence 1
versionId
}
workflow {"MegaApp flows"
name "MegaCorp"
vendor "A really expensive proprietary licence"
licence 1
versionId
} }
Important
Watch out for the UK spelling of the word licence (with a c).
Name, vendor and licence can be set to any string you like, they don't have to be Corda identities.
Target versioning is a new concept introduced in Corda 4. Learn more by reading versioning
. Setting a target version of 4 opts in to changes that might not be 100% backwards compatible, such as API semantics changes or disabling workarounds for bugs that may be in your apps, so by doing this you are promising that you have thoroughly tested your app on the new version. Using a high target version is a good idea because some features and improvements are only available to apps that opt in.
The minimum platform version is the platform version of the node that you require, so if you start using new APIs and features in Corda 4, you should set this to 4. Unfortunately Corda 3 and below do not know about this metadata and don't check it, so your app will still be loaded in such nodes and may exhibit undefined behaviour at runtime. However it's good to get in the habit of setting this properly for future releases.
Note
Whilst it's currently a convention that Corda releases have the platform version number as their major version i.e. Corda 3.3 implements platform version 3, this is not actually required and may in future not hold true. You should know the platform version of the node releases you want to target.
The new versionId
number is a version code for your app, and is unrelated to Corda's own versions. It is used to block state downgrades: when a state constraint can be satisfied by multiple attachments, the version is tracked in the ledger and cannot decrement. This ensures security fixes in CorDapps stick and can't be reversed by downgrading states to an earlier version. See "contract_non-downgrade_rule_ref
" for more information.
Split your app into contract and workflow JARs. The duplication between contract
and workflow
blocks exists because you should split your app into two separate JARs/modules, one that contains on-ledger validation code like states and contracts, and one for the rest (called by convention the "workflows" module although it can contain a lot more than just flows: services would also go here, for instance). For simplicity, here we use one JAR for both, but this is in general an anti-pattern and can result in your flow logic code being sent over the network to arbitrary third party peers, even though they don't need it.
In future, the version ID attached to the workflow JAR will also be used to help implement smoother upgrade and migration features. You may directly reference the gradle version number of your app when setting the CorDapp specific versionId identifiers if this follows the convention of always being a whole number starting from 1.
If you use the finance demo app, you should adjust your dependencies so you depend on the finance-contracts and finance-workflows artifacts from your own contract and workflow JAR respectively. Although a single finance jar still exists in Corda 4 for backwards compatibility, it should not be installed or used for updated apps. This way, only the code that needs to be on the ledger actually will be.
Step 3. Security: Upgrade your use of FinalityFlow
The previous FinalityFlow
API is insecure. It doesn't have a receive flow, so requires counterparty nodes to accept any and all signed transactions that are sent to it, without checks. It is highly recommended that existing CorDapps migrate away to the new API, as otherwise things like business network membership checks won't be reliably enforced.
This is a three step process:
- Change the flow that calls
FinalityFlow
. - Change or create the flow that will receive the finalised transaction.
- Make sure your application's minimum and target version numbers are both set to 4 (see step 2).
Upgrading a non-initiating flow
As an example, let's take a very simple flow that finalises a transaction without the involvement of a counterpart flow:
example-code/src/main/kotlin/net/corda/docs/kotlin/FinalityFlowMigration.kt
example-code/src/main/java/net/corda/docs/java/FinalityFlowMigration.java
To use the new API, this flow needs to be annotated with InitiatingFlow
and a FlowSession
to the participant(s) of the transaction must be passed to FinalityFlow
:
example-code/src/main/kotlin/net/corda/docs/kotlin/FinalityFlowMigration.kt
example-code/src/main/java/net/corda/docs/java/FinalityFlowMigration.java
If there are more than one transaction participants then a session to each one must be initiated, excluding the local party and the notary.
A responder flow has to be introduced, which will automatically run on the other participants' nodes, which will call ReceiveFinalityFlow
to record the finalised transaction:
example-code/src/main/kotlin/net/corda/docs/kotlin/FinalityFlowMigration.kt
example-code/src/main/java/net/corda/docs/java/FinalityFlowMigration.java
Note
All the nodes in your business network will need the new CorDapp, otherwise they won't know how to receive the transaction. This includes nodes which previously didn't have the old CorDapp. If a node is sent a transaction and it doesn't have the new CorDapp loaded then simply restart it with the CorDapp and the transaction will be recorded.
Upgrading an initiating flow
For flows which are already initiating counterpart flows then it's a matter of using the existing flow session. Note however, the new FinalityFlow
is inlined and so the sequence of sends and receives between the two flows will change and will be incompatible with your current flows. You can use the flow version API to write your flows in a backwards compatible manner.
Here's what an upgraded initiating flow may look like:
example-code/src/main/kotlin/net/corda/docs/kotlin/FinalityFlowMigration.kt
example-code/src/main/java/net/corda/docs/java/FinalityFlowMigration.java
For the responder flow, insert a call to ReceiveFinalityFlow
at the location where it's expecting to receive the finalised transaction. If the initiator is written in a backwards compatible way then so must the responder.
example-code/src/main/kotlin/net/corda/docs/kotlin/FinalityFlowMigration.kt
example-code/src/main/java/net/corda/docs/java/FinalityFlowMigration.java
You may already be using waitForLedgerCommit
in your responder flow for the finalised transaction to appear in the local node's vault. Now that it's calling ReceiveFinalityFlow
, which effectively does the same thing, this is no longer necessary. The call to waitForLedgerCommit
should be removed.
Step 4. Security: Upgrade your use of SwapIdentitiesFlow
The confidential_identities_ref
API is experimental in Corda 3 and remains so in Corda 4. In this release, the SwapIdentitiesFlow
has been adjusted in the same way as FinalityFlow
above, to close problems with confidential identities being injectable into a node outside of other flow context. Old code will still work, but it is recommended to adjust your call sites so a session is passed into the SwapIdentitiesFlow
.
Step 5. Possibly, adjust test code
MockNodeParameters
and functions creating it no longer use a lambda expecting a NodeConfiguration
object. Use a MockNetworkConfigOverrides
object instead. This is an API change we regret, but unfortunately in Corda 3 we accidentally exposed large amounts of the node internal code through this one API entry point. We have now insulated the test API from node internals and reduced the exposure.
If you are constructing a MockServices for testing contracts, and your contract uses the Cash contract from the finance app, you now need to explicitly add net.corda.finance.contracts
to the list of cordappPackages
. This is a part of the work to disentangle the finance app (which is really a demo app) from the Corda internals. Example:
val ledgerServices = MockServices(
"net.corda.examples.obligation", "net.corda.testing.contracts"),
listOf(
identityService = makeTestIdentityService(),"TestIdentity", "", "GB"))
initialIdentity = TestIdentity(CordaX500Name( )
becomes:
val ledgerServices = MockServices(
"net.corda.examples.obligation", "net.corda.testing.contracts", "net.corda.finance.contracts"),
listOf(
identityService = makeTestIdentityService(),"TestIdentity", "", "GB"))
initialIdentity = TestIdentity(CordaX500Name( )
You may need to use the new TestCordapp
API when testing with the node driver or mock network, especially if you decide to stick with the pre-Corda 4 FinalityFlow
API. The previous way of pulling in CorDapps into your tests (i.e. via using the cordappPackages
parameter) does not honour CorDapp versioning. The new API TestCordapp.findCordapp()
discovers the CorDapps that contain the provided packages scanning the classpath, so you have to ensure that the classpath the tests are running under contains either the CorDapp .jar
or (if using Gradle) the relevant Gradle sub-project. In the first case, the versioning information in the CorDapp .jar
file will be maintained. In the second case, the versioning information will be retrieved from the Gradle cordapp
task. For example, if you are using MockNetwork
for your tests, the following code:
val mockNetwork = MockNetwork(
"net.corda.examples.obligation", "net.corda.finance.contracts"),
cordappPackages = listOf(
notarySpecs = listOf(MockNetworkNotarySpec(notary)) )
would need to be transformed into:
val mockNetwork = MockNetwork(MockNetworkParameters(
"net.corda.businessnetworks.membership")),
cordappsForAllNodes = listOf(TestCordapp.findCordapp(
notarySpecs = listOf(MockNetworkNotarySpec(notary)) ))
Note that every package should exist in only one CorDapp, otherwise the discovery process won't be able to determine which one to use and you will most probably see an exception telling you There is more than one CorDapp containing the package
. For instance, if you have 2 CorDapps containing the packages net.corda.examples.obligation.contracts
and net.corda.examples.obligation.flows
, you will get this error if you specify the package net.corda.examples.obligation
.
Note
If you have any CorDapp code (e.g. flows/contracts/states) that is only used by the tests and located in the same test module, it won't be discovered now. You will need to move them in the main module of one of your CorDapps or create a new, separate CorDapp for them, in case you don't want this code to live inside your production CorDapps.
Step 6. Security: Add BelongsToContract annotations
In versions of the platform prior to v4, it was the responsibility of contract and flow logic to ensure that TransactionState
objects contained the correct class name of the expected contract class. If these checks were omitted, it would be possible for a malicious counterparty to construct a transaction containing e.g. a cash state governed by a commercial paper contract. The contract would see that there were no commercial paper states in a transaction and do nothing, i.e. accept.
In Corda 4 the platform takes over this responsibility from the app, if the app has a target version of 4 or higher. A state is expected to be governed by a contract that is either:
- The outer class of the state class, if the state is an inner class of a contract. This is a common design pattern.
- Annotated with
@BelongsToContract
which specifies the contract class explicitly.
Learn more by reading "implicit_constraint_types
". If an app targets Corda 3 or lower (i.e. does not specify a target version), states that point to contracts outside their package will trigger a log warning but validation will proceed.
Step 7. Learn about signature constraints and JAR signing
design/data-model-upgrades/signature-constraints
are a new data model feature introduced in Corda 4. They make it much easier to deploy application upgrades smoothly and in a decentralised manner. Signature constraints are the new default mode for CorDapps, and the act of upgrading your app to use the version 4 Gradle plugins will result in your app being automatically signed, and new states automatically using new signature constraints selected automatically based on these signing keys.
You can read more about signature constraints and what they do in api-contract-constraints
. The TransactionBuilder
class will automatically use them if your application JAR is signed. We recommend all JARs are signed. To learn how to sign your JAR files, read cordapp_build_system_signing_cordapp_jar_ref
. In dev mode, all JARs are signed by developer certificates. If a JAR that was signed with developer certificates is deployed to a production node, the node will refuse to start. Therefore to deploy apps built for Corda 4 to production you will need to generate signing keys and integrate them with the build process.
Step 8. Security: Package namespace handling
Almost no apps will be affected by these changes, but they're important to know about.
There are two improvements to how Java package protection is handled in Corda 4:
- Package sealing
- Package namespace ownership
Sealing. App isolation has been improved. Version 4 of the finance CorDapp (corda-finance.jar) is now built as a set of sealed and signed JAR files. This means classes in your own CorDapps cannot be placed under the following package namespace: net.corda.finance
In the unlikely event that you were injecting code into net.corda.finance.*
package namespaces from your own apps, you will need to move them into a new package, e.g. net/corda/finance/flows/MyClass.java
can be moved to com/company/corda/finance/flows/MyClass.java
. As a consequence your classes are no longer able to access non-public members of finance CorDapp classes.
When recompiling your JARs for Corda 4, your own apps will also become sealed, meaning other JARs cannot place classes into your own packages. This is a security upgrade that ensures package-private visibility in Java code works correctly. If other apps could define classes in your own packages, they could call package-private methods, which may not be expected by the developers.
Namespace ownership. This part is only relevant if you are joining a production compatibility zone. You may wish to contact your zone operator and request ownership of your root package namespaces (e.g. com.megacorp.*
), with the signing keys you will be using to sign your app JARs. The zone operator can then add your signing key to the network parameters, and prevent attackers defining types in your own package namespaces. Whilst this feature is optional and not strictly required, it may be helpful to block attacks at the boundaries of a Corda based application where type names may be taken "as read". You can learn more about this feature and the motivation for it by reading "design/data-model-upgrades/package-namespace-ownership
".
Step 9. Consider adding extension points to your flows
In Corda 4 it is possible for flows in one app to subclass and take over flows from another. This allows you to create generic, shared flow logic that individual users can customise at pre-agreed points (protected methods). For example, a site-specific app could be developed that causes transaction details to be converted to a PDF and sent to a particular printer. This would be an inappropriate feature to put into shared business logic, but it makes perfect sense to put into a user-specific app they developed themselves.
If your flows could benefit from being extended in this way, read "flow-overriding
" to learn more.
Step 10. Possibly update Vault state queries
Queries made on a node's vault can filter by the relevancy of those states to the node in Corda 4. As this functionality does not exist in Corda 3, apps targeting that release will continue to receive all states in any vault queries. In Corda 4, the default is to return all states in the vault, to maintain backwards compatibility. However, it may make sense to migrate queries expecting just those states relevant to the node in question to query for only relevant states. See api-vault-query.rst
for more details on how to do this. Not doing this may result in queries returning more states than expected if the node is using Observer node functionality (see "tutorial-observer-nodes.rst
").
Step 10. Explore other new features that may be useful
Corda 4 adds several new APIs that help you build applications. Why not explore:
- The new withEntityManager API for using JPA inside your flows and services.
reference_states
, that let you use an input state without consuming it.state_pointers
, that make it easier to 'point' to one state from another and follow the latest version of a linear state.